Virtuoso: Virual Machines' Distribution
- Computing Techniques Seminars
- Network Statistics and Tools at Fermilab
- C++ Design and Implementation Techniques by Example
- Virtuoso: Virual Machines' Distribution
- Inside std::Vector
- Grid Services-based Data Stream for LEAD
|Full Title:||Virtuoso: Virtual Machines' Distribution|
|Date & Time:||05 Aug 2003 at 09:00|
|Event Info:||Speaker: Peter A. Dinda
Virtual machine monitors, such as VMware, Virtual PC/Server, VM, and others, provide a powerful mechanism that can be used to greatly simplify wide-area distributed (or grid) computing. These tools allow us to lower the level of abtraction that resource providers can present to their users to the advantage of both. The abstraction is that of a raw machine with no software. The user can customize the hardware of his machine(s), install whatever operating system and software he needs, and instantiate and migrate his machine to run on whatever resources are most appropriate at any given time.
The Virtouso project is developing and extending the middleware services necessary to make this vision of distributed computing possible. This talk will concentrate on the issues and our work in the following areas:
a.. Virtual networking. We are developing software that overlays an adaptive, protocol-agnostic layer 2 virtual network that connects the user 's virtual machines, providing efficient making them appear to be on his local network.
b.. Remote devices. Users expect local devices such as cd-rom drives, floppies, and similar devices to be available on remote virtual machines. We are developing software to make this possible.
c.. Information services. We are extending our RGIS relational grid information service system to support virtual resources by adding the notions of virtual indirection and futures to its schema.
d.. Resource monitoring and prediction. We are developing techniques to efficiently measure, predict, and disseminate dynamic performance information about virtual resources in the context of our RPS system.
e.. Resource control. Providers need to be able to control resource contention among virtual machines to preseve the illusion of separate machines to users. We are exploring scheduling techniques to make this possible.